Models of mass transport in foods do require a large number of parameters, many of which are not available for food materials. Although data on effective diffusivity are abundant in literature, "true" diffusivities applying to the case of molecular diffusion are father scarce. In fact, measurements of molecular diffusivity (D) are not easy in foods and require dedicated devices with multiple cautions. Surprisingly, little concern has been paid to uncertainty accompanying diffusivity coefficients, although crucial. Classically, the unknown D function is identified by minimizing iteratively a merit function that assesses the closeness of the experimental concentration profiles to the simulated ones (method A). A different method compares the mass fluxes to the concentration gradients which bath can be estimated locally from concentration profiles (method B). Advantageously, knowledge about the concentration at the interface and the function linking D to concentration is not a requirement in the method B. The objective of the present study was to identify the requirements for the set of data so to reach a given level of uncertainty on D for both methods (A and B). The number of profiles (time resolution), the number of points per profile (space resolution) and the uncertainty on concentration were basic characteristics of the set of data. Virtual desorption/sorption kinetics were designed in order to compare the estimated value of D with a known reference. The probability density function of D was accessed numerically through the production of m virtual sets of data and applying m times the optimisation procedure. Thus, when the uncertainty on concentration was weak (< 5 %), both methods led to the same accuracy of D. For uncertainties on concentration higher than 5 %, the method B decreased accuracy of D. To increase the accurateness, it was necessary to increase the time and temporal resolution (e.g. error on D was close to the noise intensity if the numbers of co